biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[CI/Test] improve robustness of test (vllm_runner) (#5357)

Browse Source 
[CI/Test] improve robustness of test by replacing del with context manager (vllm_runner) (#5357)
This commit is contained in:
youkaichao
2024-06-08 01:59:20 -07:00
committed by GitHub
parent 9fb900f90c
commit 8ea5e44a43
28 changed files with 431 additions and 470 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
tests/models/test_marlin.py
View File

@@ -59,20 +59,16 @@ def test_models(
max_tokens: int,
num_logprobs: int,
) -> None:
marlin_model = vllm_runner(model_pair.model_marlin,
dtype=dtype,
quantization="marlin")
marlin_outputs = marlin_model.generate_greedy_logprobs(
example_prompts, max_tokens, num_logprobs)
del marlin_model
with vllm_runner(model_pair.model_marlin,
dtype=dtype,
quantization="marlin") as marlin_model:
marlin_outputs = marlin_model.generate_greedy_logprobs(
example_prompts, max_tokens, num_logprobs)
gptq_model = vllm_runner(model_pair.model_gptq,
dtype=dtype,
quantization="gptq")
gptq_outputs = gptq_model.generate_greedy_logprobs(example_prompts,
max_tokens,
num_logprobs)
del gptq_model
with vllm_runner(model_pair.model_gptq, dtype=dtype,
quantization="gptq") as gptq_model:
gptq_outputs = gptq_model.generate_greedy_logprobs(
example_prompts, max_tokens, num_logprobs)
check_logprobs_close(
outputs_0_lst=gptq_outputs,